Army Research Laboratory

Similar documents
Predicting Deformation and Strain Behavior in Circuit Board Bend Testing

Expressions for the Total Yaw Angle

Dynamic Uniaxial Tensile Loading of Vector Polymers

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT NUMBER

Viscoelastic Models for Nearly Incompressible Materials

Report Documentation Page

Analysis Comparison between CFD and FEA of an Idealized Concept V- Hull Floor Configuration in Two Dimensions. Dr. Bijan Khatib-Shahidi & Rob E.

REGENERATION OF SPENT ADSORBENTS USING ADVANCED OXIDATION (PREPRINT)

Closed-form and Numerical Reverberation and Propagation: Inclusion of Convergence Effects

General Atomics Pulsed Power Capacitor Comparison Test Report

Unitary Transformations in 3-D Vector Representation of Qutrit States

Diagonal Representation of Certain Matrices

REDUCING PEAK POWER IN AUTOMATED WEAPON LAYING

Report Documentation Page

SMA Bending. Cellular Shape Memory Structures: Experiments & Modeling N. Triantafyllidis (UM), J. Shaw (UM), D. Grummon (MSU)

Attribution Concepts for Sub-meter Resolution Ground Physics Models

Use of Wijsman's Theorem for the Ratio of Maximal Invariant Densities in Signal Detection Applications

High-Fidelity Computational Simulation of Nonlinear Fluid- Structure Interaction Problems

Super-Parameterization of Boundary Layer Roll Vortices in Tropical Cyclone Models

Design for Lifecycle Cost using Time-Dependent Reliability

Preliminary Comparison of March May Radiosonde Soundings with March May Climatological Profiles for Yuma, Arizona

Crowd Behavior Modeling in COMBAT XXI

Development and Application of Acoustic Metamaterials with Locally Resonant Microstructures

AN EMPIRICAL SHAPED CHARGE JET BREAKUP MODEL

Hot Environment Assessment Tool (HEAT) User s Guide for Apple Mobile Devices

USER S GUIDE. ESTCP Project ER

P. Kestener and A. Arneodo. Laboratoire de Physique Ecole Normale Supérieure de Lyon 46, allée d Italie Lyon cedex 07, FRANCE

AIR FORCE RESEARCH LABORATORY Directed Energy Directorate 3550 Aberdeen Ave SE AIR FORCE MATERIEL COMMAND KIRTLAND AIR FORCE BASE, NM

Babylonian resistor networks

Advanced Numerical Methods for NWP Models

Broadband matched-field source localization in the East China Sea*

HIGH-POWER SOLID-STATE LASER: LETHALITY TESTING AND MODELING

STUDY OF DETECTION LIMITS AND QUANTITATION ACCURACY USING 300 MHZ NMR

Guide to the Development of a Deterioration Rate Curve Using Condition State Inspection Data

Wavelet Spectral Finite Elements for Wave Propagation in Composite Plates

System Reliability Simulation and Optimization by Component Reliability Allocation

Ocean Acoustics Turbulence Study

Army Research Laboratory

Range-Dependent Acoustic Propagation in Shallow Water with Elastic Bottom Effects

Interim Research Performance Report (Monthly) 4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER

Convection and Shear Flow in TC Development and Intensification

Two-Dimensional Simulation of Truckee River Hydrodynamics

Metrology Experiment for Engineering Students: Platinum Resistance Temperature Detector

SW06 Shallow Water Acoustics Experiment Data Analysis

Improvements in Modeling Radiant Emission from the Interaction Between Spacecraft Emanations and the Residual Atmosphere in LEO

Super-Parameterization of Boundary Layer Roll Vortices in Tropical Cyclone Models

Quantitation and Ratio Determination of Uranium Isotopes in Water and Soil Using Inductively Coupled Plasma Mass Spectrometry (ICP-MS)

A report (dated September 20, 2011) on. scientific research carried out under Grant: FA

Parametric Models of NIR Transmission and Reflectivity Spectra for Dyed Fabrics

REPORT DOCUMENTATION PAGE

DIRECTIONAL WAVE SPECTRA USING NORMAL SPREADING FUNCTION

Direct Numerical Simulation of Aeolian Tones

Bone Tissue Mechanics

The Bernoulli equation in a moving reference frame

On Applying Point-Interval Logic to Criminal Forensics

Weak Turbulence in Ocean Waves

CRS Report for Congress

Modeling the Impact of Extreme Events on Margin Sedimentation

Contactless Mobility, Carrier Density, and Sheet Resistance Measurements on Si, GaN, and AlGaN/GaN High Electron Mobility Transistor (HEMT) Wafers

REPORT DOCUMENTATION PAGE

Thermo-Kinetic Model of Burning for Polymeric Materials

HIGH PERFORMANCE CONTROLLERS BASED ON REAL PARAMETERS TO ACCOUNT FOR PARAMETER VARIATIONS DUE TO IRON SATURATION

Predictive Model for Archaeological Resources. Marine Corps Base Quantico, Virginia John Haynes Jesse Bellavance

Scattering of Internal Gravity Waves at Finite Topography

Comparative Analysis of Flood Routing Methods

Dynamics of Droplet-Droplet and Droplet-Film Collision. C. K. Law Princeton University

Grant Number: N IP To compare obtained theoretical results with NPAL experimental data.

The Sediment Budget Calculator: A Webtool to Develop a Family of Viable Inlet and Adjacent Beach Sediment Budget Solutions

Extension of the BLT Equation to Incorporate Electromagnetic Field Propagation

REPORT DOCUMENTATION PAGE. Theoretical Study on Nano-Catalyst Burn Rate. Yoshiyuki Kawazoe (Tohoku Univ) N/A AOARD UNIT APO AP

Molecular Characterization and Proposed Taxonomic Placement of the Biosimulant 'BG'

Awell-known lecture demonstration1

Lectures on. Constitutive Modelling of Arteries. Ray Ogden

REPORT DOCUMENTATION PAGE

Modulation Instability of Spatially-Incoherent Light Beams and Pattern Formation in Incoherent Wave Systems

The Extratropical Transition of Tropical Cyclones

Vortex Rossby Waves and Hurricane Evolution in the Presence of Convection and Potential Vorticity and Hurricane Motion

Classification of Prostate Cancer Grades and T-Stages based on Tissue Elasticity Using Medical Image Analysis. Supplementary Document

Global Stability and Dynamics of Strongly Nonlinear Systems Using Koopman Operator Theory

Study of Electromagnetic Scattering From Material Object Doped Randomely WithThin Metallic Wires Using Finite Element Method

Coastal Engineering Technical Note

Heat Conduction by Molecular Dynamics Technique

Exact Solution of a Constrained. Optimization Problem in Thermoelectric Cooling

Fleet Maintenance Simulation With Insufficient Data

Tensor Visualization. CSC 7443: Scientific Information Visualization

PIPS 3.0. Pamela G. Posey NRL Code 7322 Stennis Space Center, MS Phone: Fax:

Flocculation, Optics and Turbulence in the Community Sediment Transport Model System: Application of OASIS Results

Quantum Computer Circuit Analysis and Design

VLBA IMAGING OF SOURCES AT 24 AND 43 GHZ

Swash Zone Dynamics: Modeling and Data Analysis

Understanding Near-Surface and In-cloud Turbulent Fluxes in the Coastal Stratocumulus-topped Boundary Layers

Computer Simulation of Sand Ripple Growth and Migration.

Parameterizing the Effects of Upper-Ocean Large Eddies on Air-Sea Interaction

A Radical Method for Calculating Muzzle Motion From Proximity Sensor Data

Optimizing Robotic Team Performance with Probabilistic Model Checking

Analysis of a Novel Fracture Toughness Testing Geometry

PREDICTION OF HIGH TEMPERATURE GREASE LIFE USING A DECOMPOSITION KINETIC MODEL

A GENERAL PROCEDURE TO SET UP THE DYADIC GREEN S FUNCTION OF MULTILAYER CONFORMAL STRUCTURES AND ITS APPLICATION TO MICROSTRIP ANTENNAS

NUMERICAL SOLUTIONS FOR OPTIMAL CONTROL PROBLEMS UNDER SPDE CONSTRAINTS

Modeling of Coastal Ocean Flow Fields

Transcription:

Army Research Laboratory Development and Implementation of a Transversely Isotropic Hyperelastic Constitutive Model With Two Fiber Families to Represent Anisotropic Soft Biological Tissues by Adam Sokolow and Samantha L. Wozniak ARL-CR-738 June 214 prepared by Oak Ridge Institute for Science and Education ORAU Maryland 4692 Millennium Drive, Suite 11, Belcamp, MD 2117 under contract: 112-112-99 and Bowhead Science & Technology, LLC 43 Bata Blvd., Suite K, Belcamp, MD 2117 under contract: W911QX-9-C-57 Approved for public release; distribution is unlimited.

NOTICES Disclaimers The findings in this report are not to be construed as an official Department of the Army position unless so designated by other authorized documents. Citation of manufacturer s or trade names does not constitute an official endorsement or approval of the use thereof. Destroy this report when it is no longer needed. Do not return it to the originator.

Army Research Laboratory Aberdeen Proving Ground, MD 215-566 ARL-CR-738 June 214 Development and Implementation of a Transversely Isotropic Hyperelastic Constitutive Model With Two Fiber Families to Represent Anisotropic Soft Biological Tissues Adam Sokolow Oak Ridge Institute for Science and Education Samantha L. Wozniak Bowhead Science & Technology, LLC prepared by Oak Ridge Institute for Science and Education ORAU Maryland 4692 Millennium Drive, Suite 11, Belcamp, MD 2117 under contract: 112-112-99 and Bowhead Science & Technology, LLC 43 Bata Blvd., Suite K, Belcamp, MD 2117 under contract: W911QX-9-C-57 Approved for public release; distribution is unlimited.

REPORT DOCUMENTATION PAGE Form Approved OMB No. 74-188 Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing the burden, to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (74-188), 1215 Jefferson Davis Highway, Suite 124, Arlington, VA 2222-432. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number. PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS. 1. REPORT DATE (DD-MM-YYYY) 2. REPORT TYPE 3. DATES COVERED (From - To) June 214 4. TITLE AND SUBTITLE Final Development and Implementation of a Transversely Isotropic Hyperelastic Constitutive Model With Two Fiber Families to Represent Anisotropic Soft Biological Tissues December 213 March 214 5a. CONTRACT NUMBER 112-112-99/W911QX-9-C-57 5b. GRANT NUMBER 5c. PROGRAM ELEMENT NUMBER 6. AUTHOR(S) Adam Sokolow and Samantha L. Wozniak 5d. PROJECT NUMBER 5e. TASK NUMBER 5f. WORK UNIT NUMBER 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT NUMBER ORAU Maryland 4692 Millennium Drive, Suite 11 Belcamp, MD 2117 Bowhead Science & Technology, LLC 43 Bata Blvd., Suite K Belcamp, MD 2117 ARL-CR-738 9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) U.S. Army Research Laboratory ATTN: RDRL-WMP-B Aberdeen Proving Ground, MD 215-566 1. SPONSOR/MONITOR'S ACRONYM(S) 11. SPONSOR/MONITOR'S REPORT NUMBER(S) 12. DISTRIBUTION/AVAILABILITY STATEMENT Approved for public release; distribution is unlimited. 13. SUPPLEMENTARY NOTES 14. ABSTRACT We developed a transversely isotropic hyperelastic constitutive model to capture the anisotropy of the spinal intervertebral discs. The constitutive model was implemented in the finite element code SIERRA/SolidMechanics. This report documents the continuum theory of a transversely isotropic hyperelastic constitutive model with two fiber families and verifies the implementation of the numerical model on a variety of single element tests. The algorithm used to apply this model to spinal intervertebral discs is covered in depth, and an example simulation result is shown. Application of this model to other biological tissues, such as brain tissue and skeletal muscles, is also discussed. 15. SUBJECT TERMS transversely isotropic hyperelastic, two fiber families, nearly incompressible, anisotropic, biological tissues 16. SECURITY CLASSIFICATION OF: a. REPORT b. ABSTRACT c. THIS PAGE Unclassified Unclassified Unclassified 17. LIMITATION OF ABSTRACT UU ii 18. NUMBER OF PAGES 5 19a. NAME OF RESPONSIBLE PERSON Adam Sokolow 19b. TELEPHONE NUMBER (Include area code) 41-36-2985 Standard Form 298 (Rev. 8/98) Prescribed by ANSI Std. Z39.18

Contents List of Figures v List of Tables viii Acknowledgments ix 1. Introduction 1 2. Structure and Biology of the Spine and Intervertebral Discs 2 3. Transversely Isotropic Hyperelastic Constitutive Model With Two Fiber Families 4 4. Verification of Numerical Model 1 4.1 Stretch and Compression Test for a Single Fiber Family... 11 4.2 Shear Test for a Single Fiber Family... 13 4.3 Compression Test With Two Fiber Families... 15 5. Determining the Fiber Directions for an Intervertebral Disc 16 6. Future Applications of the Model 21 6.1 Modeling Intervertebral Discs... 22 6.2 Modeling Brain Tissue... 24 6.3 Incorporating Fiber Prestresses... 27 6.4 Incorporation of Active Contractile Fiber Response... 29 iii

7. Concluding Remarks 31 8. References 32 Distribution List 35 iv

List of Figures Figure 1. Bony anatomy of the spinal column (panel a) and a typical vertebra (panel b). Vertebra are color-coded according to their location classification. Panel c is an illustration (not drawn to scale) of an intervertebral disc showing the lamellar architecture of the annulus fibrosus (white) which surrounds the nucleus pulposus (grey). Some layers have been cut away from the annulus fibrosus to show the fiber network within the lamellae. Note that only four layers of lamellae are depicted in the figure but the annulus fibrosus usually has 15 25 layers. Collagen fibers (diagonal lines) are oriented at ±3 to the transverse plane of the disc, with the direction alternating between adjacent lamellae in the annulus fibrosus.... 3 Figure 2. Single fiber family stretch test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for three sets of fiber orientation vectors a. Panel a shows the stress response of a single fiber family with initial direction vector in the Y Z-plane. Similarly, panels b and c show a fiber family with initial direction vector in the XZ-plane, and XY -plane, respectively.... 12 Figure 3. Single fiber family compression test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for three sets of fiber orientation vectors a. Panel a shows the stress response of a single fiber family with initial direction vector in the Y Z-plane. Similarly, panels b and c show a fiber family with initial direction vector in the XZ-plane, and XY -plane, respectively.... 12 Figure 4. Single fiber family shear in Y-direction test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for three sets of fiber orientation vectors a. Panel a shows the stress response of a single fiber family with initial direction vector in the Y Z-plane. Similarly, panels b and c show a fiber family with initial direction vector in the XZ-plane, and XY -plane, respectively.... 14 v

Figure 5. Two fiber families compression test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for two fiber family orientation vectors a and g... 16 Figure 6. Local coordinate system for an intervertebral disc. An idealized intervertebral disc in the reference coordinate system. The point r is on the surface of the cylinder with corresponding normal vector ˆn, chosen tangent vector ˆt, and binormal vector ˆb. The vectors a and g for this point are also shown.... 17 Figure 7. Determining fiber directions in the spine. Panel a is the meshed L 3 L 4 L 5 segment of our spine model. In this drawing, the anterior (forward facing) side corresponds to the Y -axis and the vertical is Z. The centroids of all the elements of the beige intervertebral disc are shown as blue dots in panel b. A linear fit to the top layer of centroids (cyan) gives the slope of that layer. The angle normal to the plane of the intervertebral disc can be determined from this slope (red arrow) relative to the Z-axis (black arrow). This top layer is shown after it is rotated and projected into the XY -plane in panel c. A convex hull algorithm determines the outermost ring of centroids (cyan open circles) from which a parameterization yields the binormal vector ˆb (red arrows). Using the angle normal to the plane of the intervertebral disc as the Figure 8. surface tangent vector ˆt and the binormal vector from the convex hull ˆb, the fiber orientations can be determined. Panel d shows a close-up of a wireframe of the original mesh where the calculated two fiber families are shown in red and green.... 19 Comparison of constitutive models. Panels a, c, e, g, i, k, and m summarize the simulation results for our transversely isotropic hyperelastic constitutive model with two fiber families. Panels b, d, f, h, j, l, and n are the results of that same material without a fiber response. In both cases, the specimen is subjected to a 1% compressive strain. The resultant pressure, and six components of stress are shown in false color. The orientation of the intervertebral discs relative to a lab coordinate system is shown in panel n and the color bar applies to all panels.... 23 vi

Figure 9. An approach to extract two fiber families from diffusion spectrum imaging (DSI). Panels a and c are the raw data from DSI from two exemplary voxels. The color scale represents the relative strength of the specific direction sampled. The first is a case where there is only a single material direction and the second is a case where there are two material directions. Panels b and d show the result of applying our method to approximate the directions from the raw data.... 25 vii

List of Tables Table 1. Summary of material parameters used for the purpose of verifying the numerical model.... 1 viii

Acknowledgments We would like to thank Justin McKee for providing an example voxel from the diffusion spectrum imaging used on the brain. We would also like to thank Dr. Mike Scheidler for an enlightening conversation regarding the use of reference configurations to represent prestresses. This research was supported in part by an appointment to the Postgraduate Research Participation Program at the U.S. Army Research Laboratory (ARL) administered by the Oak Ridge Institute for Science and Education through an interagency agreement between the U.S. Department of Energy and ARL. ix

INTENTIONALLY LEFT BLANK. x

1. Introduction There is an increasing interest within the automotive, military, and clinical communities to use computer simulations to predict injury in humans. Researchers have developed finite element models of biological tissues and subjected their models to various loading conditions (1 4). Finite element models rely on accurate constitutive models of these tissues and therefore researchers must make critical decisions regarding the appropriate level of detail to include. While it is safe to assume anisotropy does exist in biological tissues, the role it plays during a high loading rate event, such as a blast, is not well understood. Numerical models that can turn these effects on or off are an incredibly useful tool. The biological tissues that we consider here are soft and often have a high water content that places their bulk moduli close to that of water, i.e., around 2.3 GPa. However, they also typically have relatively low shear moduli, making them nearly incompressible. An extreme example is brain tissue where the lowest shear modulus values measured are around 2 kpa, cf. (5 1). Because these tissues have low shear moduli, they can reach extremely large shear strains making the nonlinear response of the tissues extremely important to characterize. These tissues typically are viscoelastic in nature as well; however, in this report we do not include rate-dependent behavior. Instead, we focus on the fibrous structures of soft tissues and how this structural anisotropy affects the mechanical behavior under different loading mechanisms. This report introduces an anisotropic constitutive model for modeling the intervertebral discs of the spine. Since this type of model can be used for other soft biological tissues, we compromise between a fully general model and one that is overly specialized. Thus, the model is written in such a manner that it can be easily extended to capture the anisotropy of other biological tissues, such as brain or skeletal muscle. In this way, the constitutive model can be thought of as a numerical-analytical tool for investigating the mechanical response of fibrous tissue. Section 2 introduces the background information relevant to intervertebral discs. Section 3 highlights the key points in the derivation of a transversely isotropic hyperelastic constitutive model with two fiber families. We then verify the implementation of the model in section 4. Section 5 describes an algorithm of how we incorporate the fiber directions for an intervertebral disc. Our future applications of the constitutive model as it will be applied to the spine, and could be applied to the brain and skeletal muscle, are discussed in section 6. Finally, our concluding remarks are presented in section 7. 1

2. Structure and Biology of the Spine and Intervertebral Discs The vertebral column, also known as the spine, is a bony structure comprised of vertebrae and intervertebral discs, stacked alternatively on top of each other. As seen in figure 1a, five different regions make up the spine: the cervical spine, the thoracic spine, the lumbar spine, the sacrum and the coccyx. Each individual vertebra is named by referring to the first letter of their region (cervical, thoracic or lumbar), and, starting with the most superior (highest) vertebra in that region, numbered consecutively until the most inferior (lowest) vertebra in the region has been named. Between each vertebra is a soft tissue called the intervertebral disc. The intervertebral discs play a major role in the motion of the spine by supporting compressive forces experienced during flexion (bending forward) and extension (bending backward), and resisting rotation, tension, and shear forces (11). An illustration of an intervertebral disc is shown in figure 1c. The intervertebral discs are made up of two main components: an inner gelatinous region, known as the nucleus pulposus (grey region), and an outer ring of fibrosus cartilage, known as the annulus fibrosus (white structures surrounding the grey region). The annulus fibrosus is composed of 15 25 concentric rings called lamellae (four rings are depicted in black in the figure) (12, 13). These rings consist of collagen fibers embedded within an extracellular matrix (alternating sets of diagonal lines). The orientation of the fibers varies between adjacent lamellae, alternating approximately ±3 to the transverse plane of the disc. Since the matrix of the annulus fibrosus is relatively soft, the fibers are believed to play a prominent role in the intervertebral disc s response to tensile (14) and shear loading. At the boundary between the vertebra and the intervertebral disc is a thin layer of semiporous bone, known as the vertebral endplate. The endplates of a healthy disc help absorb some of the pressure that results from mechanical loading of the spine and prevent the nucleus pulposus from bulging into the adjacent vertebra (15). Over the past 6 years, there has been a substantial effort to model the spine and its individual components (16). Computational models of the spine provide researchers with an opportunity to gain a more detailed understanding of the deformation stress state and failure of the vertebrae and intervertebral discs. Typically, the annulus fibrosus is modeled in one of two ways: as a homogeneous composite of the matrix and the fibers or as an inhomogeneous composite of the matrix reinforced by collagen fibers (17). Shirazi-Adl found that representing the annulus 2

fibrosus as an inhomogeneous composite gave more realistic results and helped provide a better understanding of the biomechanical response of the intervertebral disc (14). Spring elements, truss elements, and rebar elements oriented at ±3 to the transverse plane of the intervertebral disc have all been used by researchers to model the fibers of the annulus fibrosus (1, 18 2). a) b) SPI NOUSPROCESS LAMI NA VERTEBRALFORAMEN TRANSVERSE PROCESS CERVICAL VERTEBRAE SUPERI ORARTI CULAR PROCESS PEDI CLE BODY SUPERI ORVERTEBRAL NOTCH THORACIC VERTEBRAE SUPERI ORARTI CULAR PROCESS TRANSVERSE PROCESS I NFERI ORVERTEBRAL NOTCH BODY I NFERI ORARTI CULARPROCESS SPI NOUSPROCESS c) NUCLEUS PULPOSUS LUMBAR VERTEBRAE SACRUM COCCYGEAL VERTEBRAE LAMELLAE OF ANNULUS FIBROSUS Figure 1. Bony anatomy of the spinal column (panel a) and a typical vertebra (panel b). Vertebra are color-coded according to their location classification. Panel c is an illustration (not drawn to scale) of an intervertebral disc showing the lamellar architecture of the annulus fibrosus (white) which surrounds the nucleus pulposus (grey). Some layers have been cut away from the annulus fibrosus to show the fiber network within the lamellae. Note that only four layers of lamellae are depicted in the figure but the annulus fibrosus usually has 15 25 layers. Collagen fibers (diagonal lines) are oriented at ±3 to the transverse plane of the disc, with the direction alternating between adjacent lamellae in the annulus fibrosus. 3

3. Transversely Isotropic Hyperelastic Constitutive Model With Two Fiber Families Soft tissues that are comprised of fibrous structures, such as muscles, ligaments, tendons, intervertebral discs and the brain, often exhibit strong anisotropy along these fiber directions (21). In this section, we provide the assumptions and physical arguments necessary for deriving a constitutive model for representing a fibrous material as a nearly incompressible, transversely isotropic hyperelastic material. We largely follow the work set out by Weiss et al. (22), Holzapfel (23), Pinsky et al. (24), and Nguyen and Boyce (25). Instead of presenting a fully general model and then specializing it to our application, we introduce simplifying assumptions to tailor the derivation to our specific application. We make assumptions appropriate for a nearly incompressible, transversely isotropic hyperelastic material with up to two fiber families that do not interact with one another, nor the surrounding matrix, and whose response depends only on their stretch. Let F be the deformation gradient describing the deformation of a material relative to some reference configuration. The polar decomposition of F is given by F = RU = V R, (1) and because R is a properly orthogonal rotation matrix, the eigenvalues of F are the same as those for U and V, the right and left stretch tensors. These eigenvalues are also the principal stretches, which we denote as λ i. We define the right and left Cauchy-Green deformation tensors, C and B, respectively C F T F B F F T (2a) (2b) which have the eigenvalues λ 2 i. The ratio of the current specific volume to the reference specific volume is the Jacobian and is given by the determinant of the deformation gradient: J det F. (3) 4

The Jacobian allows us to define the distortional part of the deformation gradient: F J 1 3 F (4) The distortional part of the deformation gradient essentially normalizes the volume changes associated with the deformation and is denoted by a bar. This can be seen by taking the determinant of F : det F = det(j 1 3 F ) = (J 1 ) det(f ) = (J 1 )(J) = 1. (5) We note that the eigenvalues of F are J 1 3 λ i λ i. Analogous to the traditional approach in defining the right and left Cauchy-Green deformation tensors, we define the so-called modified right and left Cauchy-Green deformation tensors: C F T F B F F T. (6a) (6b) The eigenvalues of C and B are λ 2 i. We next consider the invariants of the deformation tensors as they will be important for deriving our hyperelastic constitutive response from an energy density function. Let I 1, I 2, and I 3 denote the first three invariants of C and B: I 1 tr C = tr B (7a) I 2 1 2 [ (tr C) 2 + tr C 2] = 1 2 [ (tr B) 2 + tr B 2] (7b) I 3 det C = det B = J 2 (7c) and I 1, I 2, and I 3 the invariants of C and B: I 1 tr C = tr B (8a) I 2 1 [ (tr ) ] 2 2 C + tr C = 1 [ (tr ) ] 2 2 B + tr B 2 2 (8b) I 3 det C = det B = 1. (8c) To incorporate anisotropy into our description, we define two fiber family directions in the reference configuration a and g, with the property that a = 1 and g = 1. The 5

corresponding deformed fibers are given by applying the deformation gradient to the fiber direction in the reference configuration so that a = F a, a = F a (9a) g = F g, g = F g. (9b) The lengths of the deformed fiber families are at a = (F a ) T F a = a T F T F a = a T Ca (1) and since a = 1, the fiber stretch λ a is λ a = at a at a = a T Ca. (11) The same arguments can be made for the second fiber family, resulting in λ g = g T Cg. (12) Physical arguments, see for example, Weiss et al. (22) or Holzapfel (23), lead to the conclusion that the energy density function must depend on an even function of the fiber directions. Thus, one can conclude that the energy density function must depend on the structure tensor a a and g g. For notational simplicity, let: A a a, and G g g (13) and in the deformed configuration, A a a, and G g g. (14) These additional tensors introduce additional pseudo-invariants of C, A, and G, which are 6

given by I 4 a T Ca = λ 2 a I 5 a T C 2 a I 6 g T Cg = λ 2 g I 7 g T C 2 g I 8 a T Cg I 9 a T g. (15a) (15b) (15c) (15d) (15e) (15f) Similar expressions can be derived for the pseudo-invariants of the distortional tensors, but for reasons that will soon be clear we only include I 4 and I 6, I 4 a T Ca = J 2 3 λ 2 a = λ 2 a I 6 g T Cg = J 2 3 λ 2 g = λ 2 g. (16a) (16b) The energy density function φ for a hyperelastic material is often written in terms of the right Cauchy-Green deformation tensor from which the second Piola-Kirchhoff stress tensor Σ can be determined, Σ = 2 φ C. (17) The energy density function can equivalently be thought of as some function of the first three invariants of either C or B. Since the eigenvalues of C and C or equivalently, B and B, are related by J, the energy density function can also be expressed in terms of the first three invariants of either C or B. Thus, the motivation for introducing the modified deformation tensors is that the energy density can be written as some function of the Jacobian J and the modified right Cauchy-Green deformation tensor C, i.e., φ = φ(j, C). (18) Equivalently, the energy density must be some properly invariant function of the nine invariants previously discussed: φ = φ(j, I 1, I 2, I 4, I 5, I 6, I 7, I 8, I 9 ), (19) where we ve intentionally isolated out the dependence of I 3 = J. Since little is known about the actual constitutive response of these types of materials, a common simplification introduced is to assume that the fibers do not interact with one another. Thus, we assume that the mechanical 7

response is proportional to the first three isotropic invariants and only the stretches of the fiber families (I 4, I 6 ), i.e., assuming the energy density is a function of fewer parameters: φ = φ (J, I 1, I 2, I 4, I 6 ). (2) An additional simplification can be made when considering nearly incompressible materials. Typically for these soft materials, the energy density is decoupled into a spherical part (relating to pressures resulting from volume change) and a deviatoric response (shear response independent of volume changes). This is only approximately true for a real material since there will be coupling of pressure and shear terms at large deformations. This assumption decouples the energy density as follows: φ = φ s (J) + φ dev (I 1, I 2, I 4, I 6 ). (21) where it can be shown that the pressure p is and the deviatoric part of the Cauchy stress tensor dev T is p = φ s J, (22) dev T = 2 [ J dev F φ ] dev C F T. (23) The partial derivative of the energy density with respect to the distortional part of the right Cauchy-Green tensor C can be expanded using the chain rule. This procedure requires the following additional partial derivatives: I 1 C = I I 2 C = I 1I C I 4 C = A I 6 C = G. (24a) (24b) (24c) (24d) Thus, the deviatoric part of the Cauchy stress can be written: dev T = 2 J dev [F (( ) φ φ + I 1 I φ C + φ A + φ ) ] G F T. (25) I 1 I 2 I 2 I 4 I 6 8

Multiplying through by F on the left and its transpose F T on the right allows us to calculate the deviatoric part of the Cauchy stress in the reference configuration dev T = 2 [( ) φ J dev φ + I 1 B φ B 2 + φ A + φ ] G. (26) I 1 I 2 I 2 I 4 I 6 Following what has been done previously in the literature, we further assume that the isotropic response is that of the well known Mooney-Rivlin model, so that φ/ I 1 = C 1 and φ/ I 2 = C 1 are constants. We also assume that the stress response of both of the fiber families follow the same functional form, i.e., φ/ I 6 = φ/ I 4 = F(I), so that the deviatoric part of the Cauchy stress is dev T = 2 [ J dev (C1 ) ] + I 1 C 1 B C1 B 2 + F(I 4 )A + F(I 6 )G. (27) As is typical for nearly incompressible materials, we assume the spherical part of the Cauchy stress to be p = κ ln J, (28) where κ is the bulk modulus. Thus, the total Cauchy stress is given by: T = κ ln JI + 2 [ J dev (C1 ) ] + I 1 C 1 B C1 B 2 + F(I 4 )A + F(I 6 )G. (29) We have intentionally avoided discussing the functional form of the fiber response and avoided writing down the hyperelastic energy density function explicitly. The choice of the fiber response depends on the biological tissue being modeled. In the literature, collagen fibers (25), as well as other fibers, have been modeled using an exponential function (22, 24). This gives a particular definition of the fiber response F: F(I) C i ( exp [ βi (I 1) ] 1 ) (3) where the values of C i and β i can depend on the fiber family. The fiber response function takes the barred invariant I of a fiber family and returns the stress that results, thus, in practice, I in equation 3 will be either I 4 or I 6. While an energy density function can be written down for the case of equation 3, it may not be possible for all cases and functional forms of F. An example of this might be including damage or dissipation to the fiber response. Section 6.3 discusses how this model can be extended to include a prestress and section 6.4 discusses an active contractile component that responds to applied strains. 9

4. Verification of Numerical Model This section briefly covers various test cases to illustrate that our constitutive model has been implemented properly and behaves as expected. We conducted four simulations on a single element where we control the input fiber direction and the imposed deformation. The first three cases involve a single fiber family, and the fourth incorporates both fiber families. To ensure a set of rigorous tests, we considered compression, extension, and shear cases for a number of fiber family orientations. While closed-form solutions to each deformation were worked out using MuPad, they are too lengthy to be of any real analytical use. Instead, we compared our simulation results directly against the theoretically predicted responses in various figures where the angle of the fiber families vary between and π. In the first three tests, the initial fiber directions were taken so that the fiber had no component in the x, y, or z axis (corresponding to the planes Y Z, XZ, and XY ), respectively, i.e., a = (, cos θ, sin θ), a = (cos θ,, sin θ), or a = (cos θ, sin θ, ) (31) By sweeping θ from to π, we tested cases for which the fiber direction was not along a principal axis of strain. Table 1 lists the material parameters used in this verification. In the single fiber family cases, C g =. These parameters were chosen because of their relevance to the soft intervertebral discs and to illustrate an important issue regarding the sensitivity of the fibers to numerical error (see shear test for a single fiber family). Additional tests (not shown here) further verified this model for additional choices of material parameters. Table 1. Summary of material parameters used for the purpose of verifying the numerical model. κ (MPa) C 1 (kpa) C 1 (kpa) C a (MPa) β a C g (MPa) β g 7.5 3 75 8 1 4 1 C g = in the single fiber family cases 1

4.1 Stretch and Compression Test for a Single Fiber Family In these first two tests, the element is either stretched or compressed in the Z direction. The results of these test are shown in figures 2 and 3, respectively. In both cases, the physical components of the deformation gradient are given by [F ] = 1 1 α, (32) so that in compression α < 1, and in tension α > 1. In both cases, α is also the value of the Jacobian. Therefore, these tests also verify that the pressure response is implemented correctly. It is important to note the fiber stretches that are predicted in each case. From equation 15a, the fourth pseudo-invariant is I 4 = a2 x + a 2 y + α 2 a 2 z, (33) α 2 3 so that in both the stretch and compression test I 4 = cos θ2 + α 2 sin θ 2, I 4 = cos θ2 + α 2 sin θ 2 α 2 3 α 2 3, or I 4 = 1 α 2 3. (34) This implies that even for the case where the fiber has no z component, the fiber response will be a constant and independent of θ since equation 3 depends only on I 4. Figure 2 shows the results for the single fiber family stretch tests where α = 1.5. The simulation results are shown as the symbols (x s) and the theoretical response as solid lines. Note that each symbol represents a separate single-element simulation with a unique fiber family orientation, so that 36 simulations are represented in each panel. Each component of the Cauchy stress T is represented by a unique color. The individual panels plot the stress as it depends on the angle θ, which specifies the undeformed fiber directions given by equation 31. The setup for figure 3 is the same, with the exception that α =.5. Even at these large deformations, both figures 2 and 3 show excellent agreement between the numerical implementation of the constitutive model and the theoretical predictions regardless of the fiber plane or angle. Depending on the available experimental data and or the qualitative features desired from the model, equation 3 could be rewritten to depend on I 4 instead of I 4 with minimal alterations to the code. 11

2 a) a =(,cosθ,sinθ) 2 b) a =(cosθ,,sinθ) 15 c) a =(cosθ,sinθ,) 15 15 1 1 1 5 Stress (MPa) 5 Stress (MPa) 5 Stress (MPa) 5 5 5 1 1 1 2 3 Angle θ 1 1 2 3 Angle θ 15 1 2 3 Angle θ Figure 2. Single fiber family stretch test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for three sets of fiber orientation vectors a. Panel a shows the stress response of a single fiber family with initial direction vector in the Y Z-plane. Similarly, panels b and c show a fiber family with initial direction vector in the XZ-plane, and XY -plane, respectively. 2 a) a =(,cosθ,sinθ) 2 b) a =(cosθ,,sinθ) 2 c) a =(cosθ,sinθ,) 15 15 15 1 1 1 Stress (MPa) 5 Stress (MPa) 5 Stress (MPa) 5 5 5 5 1 1 1 2 3 Angle θ 1 1 2 3 Angle θ 15 1 2 3 Angle θ Figure 3. Single fiber family compression test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for three sets of fiber orientation vectors a. Panel a shows the stress response of a single fiber family with initial direction vector in the Y Z-plane. Similarly, panels b and c show a fiber family with initial direction vector in the XZ-plane, and XY -plane, respectively. 12

4.2 Shear Test for a Single Fiber Family The third case considered for the single fiber family was a shearing in the Y direction in the Z-plane (figure 4). In this case, the physical components of the deformation gradient are given by [F ] = Thus the fourth pseudo-invariant takes on the form 1 1 α 1. (35) I 4 = 1 + 2α a y a z + α 2 a 2 z (36) so that in each of the cases in equation 31: I 4 = 1 + 2α cos θ sin θ + α 2 sin θ 2, I 4 = 1 + α 2 sin θ 2, or I 4 = 1. (37) Figure 4 compares the simulation results for the single fiber family in shear. Panels a and b show excellent agreement between the theoretically predicted values and the simulation results. Panel c, however, shows some noteworthy deviations from the predicted theoretical curves. The deviation from the theoretical curves is an important issue to highlight and is entirely due to the small numerical error that SIERRA introduces in the rotation matrices during integration steps. Note that the scale bars in panel c are in kpa, while the other two panels are in MPa. Also note that for this specific example (see table 1), the choice of moduli place the isotropic response three orders in magnitude smaller than the fiber response. This particular test is an extreme case where the deformation leaves the fibers unchanged so that I 4 should be unity and that by equation 3, the fiber response should be, leaving only the isotropic stress response. Upon closer inspection, the rotation matrices calculated by SIERRA had small errors when compared to the theoretically predicted values. In the pure shear case described previously, components of the left-stretch tensor V and the rotation tensor R should be as follows: [V ] = 1 2+α 2 4+α 2 α 4+α 2 α 4+α 2 2 4+α 2, [R ] = 1 2 4+α 2 α 4+α 2 α 4+α 2 2 4+α 2. (38) 13

12 1 8 a) a =(,cosθ,sinθ) 3 25 2 b) a =(cosθ,,sinθ) 5 4 c) a =(cosθ,sinθ,) Stress (MPa) 6 4 2 2 Stress (MPa) 15 1 5 5 Stress (kpa) 3 2 1 4 6 1 15 8 1 2 3 Angle θ 2 1 2 3 Angle θ 1 1 2 3 Angle θ Figure 4. Single fiber family shear in Y-direction test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for three sets of fiber orientation vectors a. Panel a shows the stress response of a single fiber family with initial direction vector in the Y Z-plane. Similarly, panels b and c show a fiber family with initial direction vector in the XZ-plane, and XY -plane, respectively. In practice, however, we found even with strict demands on convergence criteria (in the implicit case) or small time steps (in the explicit case), the yy, yz, zy, and zz components of the actual tensors deviated from the theoretically predicted values [V sim ] [R sim ] 1 2+α 2 4+α 2 ɛ α 4+α 2 + ɛ/2 α 4+α 2 + ɛ/2 2 4+α 2 + ɛ/2 1 2 4+α 2 ɛ/1 α 4+α 2 ɛ/2 α 4+α 2 + ɛ/2 2 4+α 2 + ɛ/2, (39). (4) We estimated ɛ from a few numerical simulations of the shear tests and found that its value typically was small, ɛ 1 4. The product of V and R is F, the deformation gradient, which to first order in ɛ is [F sim ] 1 1 ɛf yy (α) α ɛf yz (α) ɛf zy (α) 1 + ɛf zz (α), (41) 14

where f ij (α) are functions of the deformation. These functions were close to unity and their functional form is suppressed since they are not essential to the arguments that follow. Using this form of the deformation gradient, we can obtain an approximate expression for the fourth pseudo-invariant to the first order in ɛ: I 4 1 ɛf(α, θ). (42) Inserting this expression into equation 3 gives F(I 4 ) = C 3 [exp( β a ɛf(α, θ)) 1] ɛc 3 β a f(α, θ). (43) Using the values from table 1 and ɛ = 1 4 gives a fiber response, on the order of 1 kpa, whose magnitude will depend further on the deformation and the angle of the fiber. This is all of the same order as the error in figure 4c. Thus, the choice of an exponential function can make the simulation very sensitive to numerical error. 4.3 Compression Test With Two Fiber Families The previous simulations were performed for single fiber families. The final verification of our implementation is a two fiber family test, in which we verify that the fiber families behave properly when both families are used. For this test, the fiber families are represented by the vectors: a = (cos θ,, sin θ), and g = (, sin θ, cos θ). (44) This choice of fiber directions can be either orthogonal or non-orthogonal (a g = sin θ cos θ) depending on θ. Figure 5 compares the simulation versus theoretical results for a compression test. By varying θ, our choice of a and g sweep the fiber families through the XZ- and Y Z-planes, respectively. This figure also shows excellent agreement between the theoretically predicted response and the simulation results. 15

2 15 a =(cosθ,,sinθ) g =(,sinθ,cosθ) 1 Stress (MPa) 5 5 1 15.5 1 1.5 2 2.5 3 Angle θ Figure 5. Two fiber families compression test. Simulation (symbols) comparison against theoretical (solid lines) for the components of the Cauchy stress T xx (red), T yy (blue), T zz (black), T xy (cyan), T yz (magenta), and T zx (green) for two fiber family orientation vectors a and g. 5. Determining the Fiber Directions for an Intervertebral Disc This section discusses the overall approach of how we approximate the fiber family directions within intervertebral discs. We begin with a more mathematical description of the fiber orientations within a simplified intervertebral disc and then discuss details of an algorithm to handle some more general geometries. Our constitutive model is designed to read in a file that lists the fiber family directions as they change in space. Although this model was designed initially for intervertebral discs, it can in theory be extended to handle a number of other materials that have the same feature of one or two fiber directions, e.g., collagen fibers of the cornea, striated muscle fibers in skeletal muscles, multiple axonal directions within the brain. In each case, the choice of the fiber direction, or the manner in which it is assigned to an element, could vary drastically. Since we typically keep the same mesh from one simulation to another, we separated the calculation of the fiber directions per element into a preprocessing step handled in MATLAB. This choice avoided repeated upfront costs associated with determining where fibers were with respect to a mesh, and enables rapid changes to be made without requiring a recompile of SIERRA. 16

As discussed in section 2, the intervertebral discs are reinforced by collagen fibers in the annulus fibrosus. Figure 1c shows that these fibers typically have a regular arrangement within the lamellae of the annulus fibrosus. To describe these directions, we start by approximating the intervertebral disc as a cylinder, as shown in figure 6. Next, we consider a point on the surface of the cylinder, in the figure this is given by some vector r in the reference frame. This surface corresponds to a single lamellae ring. A surface normal can be defined on the cylinder as ˆn for which there is a tangent plane. In this example, we set the tangent vector ˆt to be parallel with the Z-axis, however, in general this will not be the case for an intervertebral disc since it will be taken to coincide with the normal to the transverse plane of the intervertebral disc. By definition, the binormal vector is ˆb = ˆt ˆn. According to the experimental literature (12, 19), the two fiber family orientations a and g are perpendicular to the surface normal vector ˆn and make an angle θ with the tangent vector, i.e., a ˆn =, a ˆt = cos θ a (45) g ˆn =, g ˆt = cos θ g (46) and θ a = θ g. (47) Z θ g r ˆt a ˆb ˆn Y X Figure 6. Local coordinate system for an intervertebral disc. An idealized intervertebral disc in the reference coordinate system. The point r is on the surface of the cylinder with corresponding normal vector ˆn, chosen tangent vector ˆt, and binormal vector ˆb. The vectors a and g for this point are also shown. 17

If, in addition to the cylinder shown in figure 6, a second concentric cylinder with a smaller radius is added, it would share the same fiber orientations. The experimental literature describes a similar concentric structure to the orientation of the fiber families in the lamellae rings within the annulus fibrosus. A more generic notion of this concept will be used later when we describe how we determine fiber directions for a specific annulus fibrosus geometry. Thus, if one can determine a surface normal and the tangent vector that is parallel to the intervertebral disc s vertical axis (normal to the transverse plane of the intervertebral disc), then one can calculate the fiber family orientations by applying the appropriate rotations. To extend the application of these mathematical concepts to a more complicated geometry using a semi-automated approach, the overall procedure is split between two programs: a command line tool and MATLAB. The first step is handled through a collection of scripts and a tool provided in the SEACAS toolbox that is included with SIERRA, namely GROPE. This command allows one to manipulate and survey the mesh through the command line and send results to an ASCII file. We used GROPE to extract the centroid location of each element within a given annulus fibrosus and saved it to a delimited file. We use this centroid data as a point-cloud approximation for the actual geometry that we manipulate in MATLAB. A mesh of the spinal segment L 3 L 4 L 5 is shown in figure 7a and a sagittal view of the corresponding point-cloud for a single annulus fibrosus is shown in figure 7b. 18

b) b) a) 15 ˆt d) Z (mm) 1 5 5 2 1 1 2 X (mm) c) 2 1 X (mm) 1 2 2 2 Y (mm) Figure 7. Determining fiber directions in the spine. Panel a is the meshed L 3 L 4 L 5 segment of our spine model. In this drawing, the anterior (forward facing) side corresponds to the Y -axis and the vertical is Z. The centroids of all the elements of the beige intervertebral disc are shown as blue dots in panel b. A linear fit to the top layer of centroids (cyan) gives the slope of that layer. The angle normal to the plane of the intervertebral disc can be determined from this slope (red arrow) relative to the Z-axis (black arrow). This top layer is shown after it is rotated and projected into the XY -plane in panel c. A convex hull algorithm determines the outermost ring of centroids (cyan open circles) from which a parameterization yields the binormal vector ˆb (red arrows). Using the angle normal to the plane of the intervertebral disc as the surface tangent vector ˆt and the binormal vector from the convex hull ˆb, the fiber orientations can be determined. Panel d shows a close-up of a wireframe of the original mesh where the calculated two fiber families are shown in red and green. The algorithm developed in MATLAB is broken into several steps. It is applied to a set of intervertebral discs, but we only discuss its application to a single intervertebral disc. The first step of the MATLAB code was to simplify the three-dimensional point cloud down to a stack of two-dimensional points. This step is guided by user input so that the centroids of elements that all belonged to the same layer of the mesh could be selected. This simplifying step was only possible because a clear sweep direction of the mesh could be defined. Figure 7b shows that a view of the XZ-plane provides enough space between layers to differentiate them. Each layer of the intervertebral disc is a collection of three-dimensional points that approximately resided in a plane. At this step we made use of another symmetry of the intervertebral discs, the L-R symmetry, which in this case is oriented with the Y -axis. A linear fit of the (X, Z) 19

coordinate points within a single layer gave an approximate slope of that plane (see figure 7b). This slope is then used to obtain a normal vector (red arrow in panel b), which typically fell within 2 of the Z-axis (black arrow). It is important to note that the term normal vector refers to the fact that the vector is normal to the plane of the single layer of element centroids. However, in our formulation, this direction actually represents the tangent vector ˆt from figure 6 since it is tangent to the surface of the annulus fibrosus. Using the slope, the layer is rotated so that the coordinates of the centroids primarily fell in the XY -plane. With the tangent vector ˆt approximately known for this layer, it remains to determine the outward normal ˆn or the binormal vector ˆb. We note that our procedure assumes that the average slope of a layer of elements provides a good estimate for determining the tangent vector ˆt. However, this will not be the case if the intervertebral disc exhibits large barreling. In this extreme case, additional considerations might be necessary to approximate ˆn or ˆt. Figure 7c shows the result of rotating and projecting the centroids of the top layer of panel b to the XY -plane. Since the centroids of the annulus fibrosus essentially form a set of elliptical rings in this plane, we use a convex hull algorithm to single out the centroids that belong to the outermost ring (cyan open circles, panel c). Using the coordinates of the outermost ring, a parametric description of the elliptical ring is formed, i.e., (X(s), Y (s)). The tangent line of the parametric curve given by (X(s), Y (s)) then represents the binormal vector ˆb, the result of this calculation is shown as the red arrows. This procedure can be repeated by eliminating the outer ring of points and reapplying a convex hull algorithm to identify the next layer of centroids. We emphasize that this algorithm heavily depends on the regularity of our mesh and that other algorithms may be necessary for more complicated meshes or geometries. The fiber orientations for the full intervertebral disc in the lab frame can be found by using a specific fiber family angle (±θ in figure 6) given relative to the tangent vector ˆt (with associated ˆb and ˆn). This is done by working in the reference frame of a single layer of elements in the intervertebral disc, i.e., the slope determined from the single layer is used to obtain X disc and the perpendicular (which corresponds to ˆt) is our Z disc. We then assign a fiber vector to coincide with Z disc, which in this frame is (,,1). The first step is rotating the fiber vector about the Y disc -axis so that it makes an angle θ with the Z disc vector (or θ depending on the fiber family). The next rotation accounts for the measured binormal vector ˆb for a given element by rotating the resultant vector about the Z disc -axis by the angles represented by the red arrows in figure 7c. The final rotation is to give the single layer of elements the appropriate slope determined earlier by the algorithm. This is accomplished by rotating the resultant vector about the Y disc -axis by the angle between the Z disc -axis and the Z-axis, an angle that was typically less than 2. The fiber vectors 2

are then translated to the element centroid location, and then translated again to return the annulus fibrosus to coincide with the reference frame. In practice, this procedure works well for objects whose cross section is convex. The results for two intervertebral discs are shown in figure 7d, which shows a magnified wireframe view of the original mesh. The fiber families are represented by two sets of vectors a (red) and g (green). While this procedure might only approximate experimental data, the strength of our approach is that we can substitute this preprocessing step with actual experimental data in the future. 6. Future Applications of the Model This section describes some applications of our transversely isotropic hyperelastic constitutive model with two fiber families. The first example, discussed in section 6.1 presents a preliminary result in modeling an intervertebral disc. The second example, section 6.2, explores the possibility of applying this model to the brain as a continuation of previous research which considered a transversely isotropic hyperelastic constitutive model with a single fiber family. The last two examples explore how the model could be generalized to incorporate a prestress (section 6.3) and an active-contractile element similar to skeletal muscle (section 6.4). Meshes were generated in Cubit (V13.1; Sandia National Laboratory). Simulations were performed using SIERRA/SolidMechanics (Adagio/Presto 4.28; Sandia National Laboratory). Adagio is an implicit, nonlinear preconditioned conjugate gradient solver and Presto is an explicit solver. Postprocessing of simulation results was carried out in ParaView (V3.14.; Kitware) and MATLAB (The MathWorks, Natick, MA). Preprocessing the fiber directions was performed using GROPE (SEACAS Toolkit, Sandia National Laboratory), and MATLAB. Additional theoretical calculations were performed using MuPad an application package of MATLAB. 21

6.1 Modeling Intervertebral Discs This section briefly covers an example application of our constitutive model applied to the intervertebral discs and compares it against an isotropic Mooney-Rivlin material. We used the algorithm described in section 5 to approximate fiber family directions for a single intervertebral disc surrounded by two vertebrae. In this geometry, the intervertebral disc is tied to the vertebrae. We quasi-statically imposed a displacement to the top vertebra equating to a 1% compressive strain, and fixed the bottom vertebra. This produced a slight barreling, or radial bulging of the intervertebral disc. The results of our simulations are shown in figure 8. For the transversely isotropic hyperelastic model with two fiber families, we used the material parameters from table 1. The same values were used for the Mooney-Rivlin material with the exception that the fibers were turned off, i.e., C a = C g =. For both material models, the annulus fibrosus was assumed to have a density of 12 kg/m 3 and the nucleus pulposus had a density of 1 kg/m 3. 22

Pressure Txx Tyy Tzz Txy Tyz e) g) i) k) str_xx 1.5e+6 f) -1.5e+6 str_xx 1.5e+6 h) -1.5e+6 str_xx 1.5e+6 j) -1.5e+6 str_xx 1.5e+6 l) -1.5e+6-1.5e+6 str_xx -1.5e+6 1.5e+6 m) n) str_zx z 1.5e+6-1.5e+6-1.5e+6 x y -1.5e+6-1.5 MPa -1.5e+6 Figure 8. Comparison of constitutive models. Panels a, c, e, g, i, k, and m summarize the simulation results for our transversely isotropic hyperelastic constitutive model with two fiber families. Panels b, d, f, h, j, l, and n are the results of that same material without a fiber response. In both cases, the specimen is subjected to a 1% compressive strain. The resultant pressure, and six components of stress are shown in false color. The orientation of the intervertebral discs relative to a lab coordinate system is shown in panel n and the color bar applies to all panels. Tzx str_zx Transversely Isotropic Model Mooney-Rivlin pressure Model With Two Fiber 1.5e+6 Families a) b) 1.5 MPa pressure 1.5e+6 str_xx -1.5e+6 1.5e+6 c) d) str_xx 1.5e+6-1.5e+6 str_yy 1.5e+6-1.5e+6 str_zz 1.5e+6-1.5e+6 str_xy 1.5e+6-1.5e+6 str_yz 1.5e+6 23